Using Domain Information for Word Sense Disambiguation

Using Domain Information for Word Sense Disambiguation Bernardo Magnini, Carlo Strapparava, Giovanni Pezzulo and Alfio Gliozzo ITC-irst, Istituto per la Ricerca Scientifica e Tecnologica, I-38050 Trento, ITALY email: { magnini, strappa, pezzulo, gliozzo }@itc.it Abstract The major goal in ITC-irst's participation at SENSEVAL-2 was to test the role of domain in- formation in word sense disambiguation. The underlying working hypothesis is that domain labels, such as MEDICINE, ARCHITECTURE and SPORT provide a natural way to establish se- mantic relations among word senses, which can be profitably used during the disambiguation process. For each task in which we participated (i.e. English all words, English 'lexical sample' and Italian 'lexical sample') a different mix of knowledge based and statistical techniques were implemented. 1 Introduction Current investigation in Word Sense Disam- biguation (WSD) at ITC-irst focuses on the role of domain information. The hypothesis is that domain labels (such as MEDICINE, ARCHITEC- TURE and SPORT) provide a natural and pow- erful way to establish semantic relations among word senses, which can be profitably used dur- ing the disambiguation process. In particular, domains constitute a fundamental feature of text coherence, such that word senses occurring in a coherent portion of text tend to maximize domain similarity. The importance of domain information in WSD has been remarked in sev- eral works, including (Gonzalo et al., 1998) and (Buitelaar and Sacaleanu, 2001). In (Magnini and Strapparava, 2000) we introduced "Word Domain Disambiguation" (WDD) as a variant of WSD where for each word in a text a domain label (among those allowed by the word) has to be chosen instead of a sense label. We also ar- gued that WDD can be applied to disambigua- tion tasks that do not require fine grained sense distinctions, such as information retrieval and content-based user modeling. For SENSEVAL- 111 2 the goal was to evaluate the role of domain information in WSD: no other syntactic or se- mantic information has been used (e.g. seman- tic relations in WoRDNET) except domain la- bels. Three systems have been implemented, in- tegrating knowledge-based and statistical tech- niques, for the three tasks we participated in, i.e. English 'all words', English 'lexical sample' and Italian 'lexical sample'. The main lexical resource for domains is "WordNet Domains", an extension of English Wordnet 1.6 (Fellbaum, 1998) developed at ITC-irst, where synsets have been annotated with domain information. 2 WordN et Domains The basic lexical resource we used in SENSEVAL- 2 is "WordNet Domains", an extension of WoRDNET 1.6 where each synset has been an- notated with at least one domain label, se- lected from a set of about two hundred la- bels hierarchically organized (see (Magnini and Cavaglia, 2000) for the annotation methodol- ogy and for the evaluation of the resource). The information from the domains that we added is complementary to what is already in WoRDNET. First of all a domain may in- clude synsets of different syntactic categories: for instance MEDICINE groups together senses from Nouns, such as doctor#i and hospi tal#i, and from Verbs …

Content truncated to 500 words